Migrant semipalmated sandpipers (Calidris pusilla) have over four decades steadily shifted towards safer stopover locations

Peregrine falcons (Falco peregrinus) have undergone a steady hemisphere-wide recovery since the ban on DDT in 1972, resulting in an ongoing increase in the level of danger posed for migrant birds, such as Arctic-breeding sandpipers. We anticipate that in response migrant semipalmated sandpipers (Calidris pusilla) have adjusted migratory behaviour, including a shift in stopover site usage towards locations offering greater safety from falcon predation. We assessed semipalmated sandpiper stopover usage within the Atlantic Canada Shorebird Survey dataset. Based on 3,030 surveys (totalling ∼32M birds) made during southward migration, 1974 - 2017, at 198 stopover locations, we assessed the spatial distribution of site usage in each year (with a ‘priority matching distribution’ index, PMD) index in relation to the size (intertidal area) and danger (proportion of the intertidal area within 150m of the shoreline) of each location. The PMD index value is > 1 when usage is concentrated at dangerous locations, 1.0 when usage matches location size, and < 1 when usage is concentrated at safer locations. A large majority of migrants are found at the safest sties in all years, however our analysis of the PMD demonstrated that the fraction increased over time. In 1974, 80% of birds were found at the safest 20% of the sites, while in 2017, this had increased to 97%. A sensitivity analysis shows that the shift was made specifically towards safer (and not just larger) sites. The shift as measured by a PMD index decline cannot be accounted for by possible biases inherent in the data set. We conclude that the data support the prediction that increasing predator danger has induced a shift by southbound migrant semipalmated sandpipers to safer sites.


INTRODUCTION
Prey animals often react to cues of predator presence with changes in morphology (Domenici et al.,   (Pomeroy, 2006;Dekker and Ydenberg, 2004;Pomeroy et al., 2008). The sensitivity analyses reported in 150 Figure S2 vary both the 150m danger distance, and the 2500m radius. 151 We calculated the danger index from the CanVec map layers data set produced by Natural Resources 152 Canada (acquired from:www.GeoGratis.gc.ca), which shows intertidal habitat and shoreline to a 153 scale of 1:50,000. We extracted a polygon of intertidal as the waterbody features labelled as 'temporary' 154 under the "Hydro" feature category within the CanVec dataset. We also extracted the highwater line layer 155 and created a buffer of 150m around that line, which was then clipped to the intertidal layer. For each 156 Universal Transverse Mercator (UTM) region, we transformed each polygon layer from original geographic   162 We describe the distribution of sandpipers across locations in each year using a 'Priority Matching 163 matches various distribution possibilities, ranging from sandpipers aggregating at dangerous locations, to 165 spreading evenly over locations, to aggregating at safer locations (Ydenberg et al., 2017). 166 The PMD index is calculated as follows. The (mean, across all surveys) number of sandpipers censused 167 ('usage') at location i in a given year is denoted U i . The total area of intertidal habitat at that location is 168 denoted A i . The safety index for the site, y i is the proportion of the site's total intertidal area that lies more 169 than 150m from the shoreline ( Figure 3B).

170
For each location, we calculated the proportional area for that location in relation to the total area surveyed 171 for all locations sampled in a given year (p i ), and proportion of the total bird usage (q i ) in a given year at 172 each location i.

173
Within each year, sites are ordered from most dangerous to safest (i.e. lowest to highest safety index) so 174 that the cumulative proportion of total area surveyed up to location i is where the cumulative proportional area of all k sites surveyed in a year cA k = 1. Analogously, the 176 cumulative proportion of usage up to location i is calculated as where i is a given location and i − 1 is the next most dangerous location. For bird usage the area under the 182 distribution is calculated as We used the trapezoid function because its estimate lies between that generated by the 'upper-step' and 184 'lower-step' functions. Sensitivity analyses using these step functions in place of the trapezoidal function 185 produce only minor differences in the results.

186
The Priority Matching Distribution index is calculated as Values of the PMD index vary systematically with the distribution of sandpipers across locations, as 188 summarized in Table 1 and shown in Figure S1.  index with and without each location ('leverage', see Table 2). Based on this, we excluded from the analysis locations (see Figure S2). We also excluded 1995, which had an extremely high count at a site surveyed in 197 no other year that had a strong influence on the annual PMD. Our final data set included 3,030 surveys 198 at 198 stopover locations, made 1974-2017 (excluding 1990, 1991, 1995, 1998, 2008, 2010, 2011, 2013, 199 2014). quadratic term, and a model with the log of the interannual trend. We assessed the fit to the linear trend by bootstrapping the original count data to compute 95% confidence intervals of the intercept and interannual 208 trend estimates.

209
Chronological trends in site survey dates or danger indices could bias our results. We examined the 210 sensitivity of the results to the temporal distribution of these factors to explore this possibility. To explore 211 the assumptions behind the PMD we modified the estimation of the index by changing the radius around 212 each point used to calculate A i from 2.5km to 1km and 5km, and modifying the distance from shore that 213 was classified as dangerous from 150m to 50m, 300m, and 450m. We also expanded the dates of surveys 214 to include the 60th, 90th, 95th, 98th quantiles of dates, and with all surveys between July and October.

215
To separate out the potential that any shift was driven by site size and not site safety, we modified the   at the safest 20% of locations (see Table 1).

241
Linear and log-linear models both estimate a decline in the PMD index over years (Table 3)   to the data.

252
The sensitivity analyses demonstrate that the variation in the number or danger levels of the sites surveyed 253 each year do not bias the PMD index estimates, and are therefore unlikely to explain the interannual trend.

254
Likewise, the modifications to the assumptions governing either the selection of data or underlying the 255 PMD calculation do not alter the results ( Figure S2). Simulating a response to sea-level rise (i.e. sites 256 becoming smaller), or limiting the counts to the peak of migration (20-60th quantile of counts) erases the 257 interannual trend in the PMD. Importantly, when locations are ranked 'small to large' rather than 'safe to

DISCUSSION
Our results show that semipalmated sandpipers aggregate at the safest migratory stopovers, and that this 265 distribution has shifted since 1974 towards stronger aggregation at safer sites. Extensive sensitivity analyses 266 establish that this shift is specifically toward sites of higher safety rather than larger size. Further, the shift accumulated that they could be captured by gulls (Lank, 1983). But as reported for western sandpipers in Aggregating in large groups also has the benefits of reducing the likelihood of being selected by a predator 293 (dilution) and increased detection of predator attacks (many eyes Roberts, 1996;Bednekoff and Lima, 294 1998;Fernández-Juricic et al., 2007;Pays et al., 2013). With predation dilution can also come increased 295 competition during foraging (Stillman et al., 1997;Vahl et al., 2005;Minderman et al., 2006). While the

CONFLICT OF INTEREST STATEMENT
The authors declare that the research was conducted in the absence of any commercial or financial 344 relationships that could be construed as a potential conflict of interest. Rapid population decline in migratory shorebirds relying on Yellow Sea tidal mudflats as stopover sites.

565
Winter body mass and over-ocean flocking as components of danger management by Pacific dunlins.